Feature Learning for Interpretable, Performant Decision Trees